Name: Atlas 2024
Brand: Boston Dynamics

Atlas 2024

Atlas (2024) by Boston Dynamics is the fully electric research-phase humanoid, featuring custom electric actuators, 360° joint rotation, and NVIDIA Jetson Thor AI for industrial task learning.

Made by

Boston Dynamics, Inc.

Software Type

Closed Source

Software Package

Integration with Boston Dynamics’ Orbit fleet management platform.

...

Report

Actuators

Fully electric actuators.

Compute

NVIDIA Jetson Thor, announced for Atlas in March 2025.

Sensors

Max Op. time

mins

Recent Robot Videos

*Aparobot claims no ownership of videos posted unless otherwise stated.

Description text

Date

Robot Brief

Atlas (2024) is the fifth-generation fully electric humanoid robot developed by Boston Dynamics, unveiled on 17 April 2024 as the immediate successor to the hydraulic HD Atlas platform, which was retired the previous day after more than a decade of research service. The 2024 electric Atlas represents a fundamental architectural departure from its predecessor: all hydraulic actuation was replaced with custom electric actuators, enabling a wider range of motion than any prior Atlas generation including 360-degree rotation at the hip, waist, and neck joints, capabilities that exceed the human range of motion. The 2024 Atlas served as Boston Dynamics' pre-production research and development platform throughout 2024 and 2025, during which time it demonstrated industrial task learning, parts sequencing in Hyundai automotive factories, reinforcement learning-based locomotion, breakdancing, backflips, and human motion-capture-trained gaits. It was superseded by the production-ready Atlas (2026) unveiled at CES on 5 January 2026.

Use Cases

Industrial Task Learning and Demonstration:
Demonstrated autonomous parts sequencing, machine tending, and object manipulation in Hyundai automotive factory environments, serving as the primary platform for validating industrial AI task workflows ahead of the 2026 commercial launch.
Reinforcement Learning Locomotion Development:
Served as the primary hardware platform for developing and validating reinforcement learning-based locomotion policies, gait training via human motion capture, and whole-body control algorithms in NVIDIA Isaac Lab.
Dynamic Athletic Motion:
Demonstrated a full repertoire of high-agility motions including running, cartwheels, breakdancing, handstands, backflips, and parkour, validating the mechanical capability of the fully electric actuation system under extreme dynamic conditions.
AI Foundation Model Integration Research:
Served as an early adopter platform for NVIDIA's Isaac GR00T framework and Jetson Thor compute integration, advancing multimodal AI perception, decision-making, and control capabilities in partnership with NVIDIA.
Human-Robot Collaboration Prototyping:
Tested safe co-working behaviours, fenceless guarding systems, and human detection protocols in preparation for industrial deployment standards in the 2026 production version.

Industries

Research and Development:
The primary purpose of the 2024 Atlas throughout its operational period. Served as Boston Dynamics' internal R&D platform for full-body electric actuation, AI-driven locomotion, and industrial task learning.
Automotive Manufacturing:
Deployed in Hyundai factory environments for parts sequencing trials and assembly task learning, providing the operational data and validated workflows that informed the 2026 Atlas production design.

Specifications

Length

Width

Height (ResT)

Height (Stand)

Height (Min)

Height (Max)

Weight (With Batt.)

Weight (NO Batt.)

Max Step Height

Max Slope

+/-

Op. Temp (min)

°C

Op. Temp (Max)

°C

Ingress Rating

No items found.

Intro

Atlas (2024) was a research platform, succeeded by the product version Atlas (2026), unveiled at CES in January 2026. Boston Dynamics did not publish a specification sheet for the 2024 research-phase electric Atlas. The robot's defining architectural feature is its custom high-powered electric actuators, manufactured using advanced 3D printing in titanium and aluminium, which replace the hydraulic system entirely and deliver components that are simultaneously lighter and stronger than their hydraulic counterparts. These actuators enable joints to rotate beyond human range of motion, with 360-degree rotation confirmed at the hip, waist, and neck. Boston Dynamics did not publish a degrees of freedom count for this version; the 56 DoF figure applies only to Atlas (2026). The actuator design achieves 85 to 90% electrical-to-mechanical efficiency and sub-10 cm foot-placement precision across the locomotion system.

The AI compute platform is confirmed as NVIDIA Jetson Thor, announced via official Boston Dynamics press release on 18 March 2025. Jetson Thor runs complex multimodal AI models that work alongside Boston Dynamics' whole-body control and manipulation controller stack. The robot employs transformer-based perception and decision modules, with locomotion and manipulation policies developed and trained using NVIDIA Isaac Lab. Boston Dynamics did not publish the sensor suite for this version. Battery specifications, battery runtime, payload, and precise walking speed were not officially published by Boston Dynamics for the 2024 research Atlas. The robot was not commercially sold during its operational period from April 2024 through January 2026.

Connectivity

WiFi connectivity for data and control
Integration with Boston Dynamics' Orbit fleet management and data analysis platform.

‍

Capabilities

360-Degree Joint Rotation:
Custom electric actuators at the hip, waist, and neck provide full 360-degree rotational freedom, enabling movement configurations that are physically impossible for a human skeleton, including getting up from the ground in non-human orientations.
High-Efficiency Electric Actuation:
Custom titanium and aluminium 3D-printed actuators deliver 85 to 90% electrical-to-mechanical efficiency, matching the force density of the prior hydraulic system while eliminating fluid maintenance and reducing operational noise.
Multimodal AI Perception:
NVIDIA Jetson Thor compute runs complex multimodal AI models integrating visual, depth, and inertial sensor streams for real-time state estimation, environmental mapping, and adaptive motion planning.
Reinforcement Learning Locomotion:
Locomotion controllers developed and trained in NVIDIA Isaac Lab using simulation-to-real transfer, enabling smooth gait transitions, dynamic recovery from disturbances, and human-motion-capture-inspired walking.
Industrial Task Sequencing:
Demonstrated autonomous parts sequencing in Hyundai factory environments, including picking, orienting, and placing unsorted automotive components onto assembly dollies in the correct order for production line use.
Self-Righting from Non-Human Orientations:
Can recover from falls and return to an upright working stance through non-human-biomechanical recovery sequences enabled by the 360-degree joint architecture.